Nitrogen (N) fertilization in corn is often based on uniform rates and yield goals without considering the spatial and temporal variability of yield potential. It is well documented how uniform N rates lead to low N use efficiencies and environmental issues, resulting in reduced profit for farmers. Several site-specific approaches have been proposed to capture the yield spatial variability and adjust N rates to the actual crop nutrient requirements. The current study presents two original, site-specific N fertilization approaches, where two approaches at integrating crop simulation models, seasonal forecast and proximal sensing were tested across two corn seasons (2019 and 2020) in a field with significant spatial variability. In the first approach, top dressing N prescription maps were determined using the DSSAT crop model run with historical weather data, while in the second one, the maps were determined coupling DSSAT with seasonal forecasts and proximal sensing. Compared to the uniform fertilization treatment, both model-based approaches led to higher yields, N efficiency and gross margin in 2019 but not in 2020. The 2020 season was characterized by several major rainfall events, which were not present in the historical or seasonal forecast datasets. This inconsistency led to a substantial underestimation of the N leaching events in both model-based methodologies and consequently to higher-than-needed N fertilizer recommendations. Future studies should therefore focus on identifying ways to provide accurate seasonal estimates of extreme weather events to enable crop models to provide better N recommendations. In addition, the integration of proximal and remote sensing data into the crop model should be tested later in the season when spatial variability in crop N status peaks. •Different weather datasets should be used to run a crop simulation model for N recommendations.•Historical and Seasonal forecasts did not consistently represent the timing and amounts of major rainfall events.•Proximal sensing should be integrated into crop models when spatial variability has its peak (V7 to V10 stages).•Autocalibration approaches for user-independent model calibrations should be integrated into the methodology.